Hello, Alexander!
Thanks for your investigation!
I was able to reproduce it locally, not easily, but possible.
It fails with the state then we have 2 indexes
(leaf_part_rri->ri_NumIndices=2) and EACH OF them have an ancestor. So,
both of them are added as arbiters.
Not sure if it correct behavior or something is broken in another place.
I'll try to dive deeper a little bit later.
Best regards,
Mikhail.